A comparison of two methods for detecting hormone peaks: The effect of sampling interval on gonadotropin peak frequency

There is no consensus on the optimum method to identify gonadotropin pulses in serum. We compared two approaches for detecting gonadotropin peaks. The first employed the conventional criterion of an increment from nadir to peak of 3 times the intraassay coefficient of variation (3 CV). The second identified peaks by Student's t test to quadruplicate measurements at each time point. We obtained blood samples every 5 min for 6 h from four women in the follicular phase. We also constructed control or noise series by subdividing single serum pools into consecutively labeled aliquots. Any variations in hormone concentration in the noise series that were identified as peaks were, by definition, false positive peaks. We evaluated the effect of sampling interval on gonadotropin peak detection by omitting data to simulate sampling every 10, 15, or 20 min. The 3 CV approach identified numerous false positive peaks in the noise series and detected as many peaks in the noise series as it did in the patient series. Increasing the sampling frequency from every 20 to every 5 min nearly doubled the apparent peak frequencies in both the patient and the noise series (P < 0.025). By contrast, the t test method detected far fewer false positive peaks and significantly more peaks in the patient series than in the noise series. Increasing the sampling frequency from every 20 to every 5 min resulted in a 50-75% increase in peak frequency by the t test method. This increase in peak frequency appeared to result from improved detection of small peaks, because samples were obtained nearer the true peaks and nadirs. The resulting increase in the nadir to peak increment made it more likely that a small peak would achieve statistical significance. We conclude that increasingly stringent criteria for pulse detection should be applied as one increases the sampling frequency, and that the t test approach is a more valid method than the 3 CV approach because it yields significantly fewer false positive peaks.